基于改进DenseNet网络的肺炎X光图像识别算法

Pneumonia X-Ray Image Recognition Algorithm Based on Improved DenseNet Network

肺炎是一种由微生物感染引起的疾病,严重时可危及生命。目前,世界上最常用的肺炎检测方法是胸部X光图像。利用深度学习图像处理算法对肺炎X光图像特征学习,可以为放射科医生临床诊断提供客观辅助。为提升深度学习DenseNet模型应用于X光图像中检测肺炎的效果,在DenseNet深度模型的基础上,在全连接层加入中心损失(Center loss),在最后输出部分将交叉熵损失函数替换为Focal-loss,提出一种改进的DenseNet模型算法。实验验证了其最高检测分类精度达到94.52%,测试精度达到90.46%。提出的算法模型有效提高了肺炎X光图像检测分类精度。

Pneumonia is a disease caused by microbial infections,which can be life-threatening in severe cases.The most commonly used method of detecting pneumonia in the world is to use chest X-ray images.Deep learning image processing algorithms are used to learn features of pneumonia X-ray images and provide objective auxiliary diagnosis for radiologists in clinical diagnosis.In order to improve the effect of deep learning DenseNet model used in X-ray images to detect pneumonia,based on the DenseNet deep model,this paper adds center loss in the fully connected layer,and replaces the cross-entropy loss function with Focal-loss in the final output part,and proposes an improved DenseNet model algorithm.Experiments verify that the highest detection and classification accuracy reaches 94.52%,and the test accuracy is 90.46%.The algorithm model proposed in this paper effectively improves the classification accuracy of pneumonia X-ray image detection.